PROJECT SUMMARY Tuberculosis (TB) requires the simultaneous administration of multiple antibiotics to eradicate heterogeneous bacterial populations. Treatment duration ranges from 6 months for drug susceptible TB to 24 months and longer for extensively resistant TB. With a number of recent drug approvals and promising clinical development candidates, there is hope for much needed treatment shortening. However, we need predictive methods to rank the very large number of possible drug combinations and reduce them to a feasible number for testing in clinical trials. Currently, drug regimens are prioritized based on efficacy in the mouse model, which despite its ease of use, is available for only a small subset of all possible combinations. In addition, differentially drug susceptible bacterial subpopulations that are found in human pulmonary lesions are not well recapitulated in murine lungs. A hallmark of TB is the formation of lesions and the coincident remarkable ability of Mycobacterium tuberculosis to persist in a variety of lesion types during drug treatment. These hard-to-treat bacterial subpopulations cause disease persistence and relapse. Therefore, key to prioritizing new regimens is systematic, high-quality in vitro measurement of multidrug regimen potencies and a framework that links in vitro measurements to efficacy in different types of human-like lesions. To do so requires in vitro models that capture key lesion-specific stressors and harness the potential of combination therapies to identify drugs that act synergistically. We propose to fill this gap by developing a data-driven pipeline to rapidly prioritize drug regimens by combining in vitro and in vivo measurements of drug action with mathematical modeling. (1) We will generate potency measurements of drug combinations under a variety of growth conditions for direct comparison with combination drug effects in lesions. (2) We will leverage the human-like properties of rabbit pathology to query drug efficacy in distinct lesion compartments. (3) We will apply the power of multiscale (molecular, cellular, granuloma and organ scales) mathematical modeling to identify the stressors that are most predictive of in vivo efficacy. To build the pipeline, we will leverage a new drug regimen that has performed surprisingly well in clinical trials but the components of which antagonize in standard potency assays in vitro: the NiX-TB regimen comprising bedaquiline-pretomanid- linezolid. Once validated for NiX-TB versus standard of care, the pipeline will be used to rationally optimize and re-invent the NiX regimen using data-driven computational simulation.